Start date: 01.01.2026

Funded by: VDI/VDE Innovation + Technology (within the Bavarian Collaborative Research Program, BayVFP)

Local head of project:

Tamara Krafft

Fabian Stieler

Abstract

Emergency medical teams are often confronted with high-risk situations in which a patient’s condition can deteriorate within minutes. While established Early Warning Scores (EWS) perform reliably in closely monitored hospital environments, they are far less effective in the prehospital setting. Limited access to vital parameters, motion-related measurement artifacts, and occasional sensor failures make accurate risk assessment challenging and have so far prevented widespread adoption of such systems in emergency medical services.

The EWAI (Early Warning via Artificial Intelligence) project aims to close this gap by developing an innovative AI-driven early warning system for the automated assessment of patient status during emergency missions. By leveraging data-driven machine learning approaches, the project seeks to identify critical clinical trajectories and specific conditions earlier and more robustly than is currently possible.

A central innovation of the project is the creation of an intelligent data platform that integrates potentially incomplete prehospital data with high-quality in-hospital records and confirmed diagnoses. By combining information across the full continuum of care, the AI system can learn from the entire treatment pathway. To address potential sensor failures in the ambulance environment, additional methods are being developed to estimate missing values using calculated surrogate parameters.

Beyond technical performance, the project places strong emphasis on the development of trustworthy AI. Explainable AI (XAI) techniques are incorporated to ensure that the system’s predictions are transparent and clinically interpretable for healthcare professionals. In doing so, EWAI contributes meaningfully to the digital transformation of emergency medicine and to improved patient safety.

Start date: 01.01.2020

Funded by: BMBF (Federal Ministry of Education and Research)

Local head of project:

Fabian Rabe

Fabian Stieler

Abstract

For years, an increasing digitization in medicine has led to an increasing amount of data. Much of this data is used for the diagnosis of diseases. Based on existing labeled data, artificial intelligence could make predictions or diagnostic suggestions for new data. However, for reliable prediction, especially using Deep Learning, a sufficiently large base of annotated data is needed, which also contains rare diagnoses in enough quantity. This is either very time- and cost-intensive or leads to the fact that existing data cannot be used.

The LIFEDATA project will create an open-source framework to address this problem. The scientific concept combines Active Learning (AL) with Deep Neural Networks (DNN) to autonomously select data points with the greatest information value (e.g. rare diagnoses). Human experts annotate these and thus enabling efficient training of machine learning models. Additionally, instead of manually labeling all data points in a dataset, semi-supervised learning combined with AL can generate many annotations semi-automatically.

Two use cases from the life sciences that have been studied in the project investigate the added value for different problems and data types. Algorithms to explain the model classifications make the decisions of the trained ML models more comprehensible.

To achieve these goals, the AL part of the framework is first implemented and evaluated through a use case with an already annotated medical image dataset. To validate the usability, universality and applicability of the annotation tool, a second use case with ECG data will be realized. Thus, in the synergy of growing data sets and technological innovation, the results of the project can lead to an improvement in patient care.

Start date: 01.03.2016

Funded by: University of Augsburg

Local head of project:  Julia Rauscher

Abstract

Medical Information Sciences as an aspect of the interdisciplinary integration of medical, biological and genetic information with special consideration of environmental influences for human health and the course of disease using Internet of Things of medical devices.

Start date: 01.03.2016

Funded by: Universität Augsburg

Local head of project:  Julia Rauscher

Abstract

Through cooperation between the Central Hospital of Augsburg and the University of Augsburg, common characteristics of long-term survivors with stomach pancreatic tumors are identified. This is intended to create a basis of offering optimal treatment in the future and thus sustainably improving the quality of life.

Start date: 01.09.2016

Funded by: BMBF (Federal Ministry of Education and Research)

Local head of project:

Bernhard Bauer

Melanie Langermeier

Fabian Rabe

Abstract

Within the funding concept “Medical Informatics” of the Federal Ministry of Education and Research (BMBF), the University of Augsburg is a partner in the DIFUTURE consortium (The Munich-Tübigen Alliance for Data Integration and Future Medicine). Further consortium partners are the Technical University of Munich, the Ludwig-Maximilian University of Munich and the Eberhard Karls University in Tübigen. The goal of this project is to improve research opportunities and patient care through the exchange and shared use of clinical and research data across locations and institutions. 

DIFUTURE Website

Start date: 01.09.2011

Funded by: University of Augsburg

Local head of project: Heiner Oberkampf

Dissertation: